The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for processing untrusted data from external web pages, which is a potential surface for indirect prompt injection.
Ingestion points: Web pages inspected via take_snapshot, evaluate_script, and DOM analysis.
Boundary markers: Instructions explicitly restrict the agent to task-relevant pages and mandate respecting network-header redaction boundaries.
Capability inventory: Tools for executing scripts in the browser (evaluate_script), capturing visual evidence, and writing large outputs to local paths via filePath are mentioned.
Sanitization: The skill depends on tool-level redaction and does not implement custom filtering for page content.
[COMMAND_EXECUTION]: The troubleshooting process involves interacting with local system resources to verify browser attachment.
Evidence: Guidelines instruct the agent to query the Chromium debugging endpoint at http://127.0.0.1:9222 and read logs from $XDG_CACHE_HOME/chrome-devtools-mcp/logs/ or ~/.cache/chrome-devtools-mcp/logs/.
[SAFE]: No malicious patterns, obfuscation, or unauthorized exfiltration attempts were detected. The skill emphasizes best practices for privacy and concurrent browser usage.

chrome-devtools